Codex Black Edition supports voice in and out. For speech-to-text you can use ElevenLabs Scribe v1, Deepgram Nova-3, local Whisper (offline and keyless), Anthropic streaming STT, or the WebSpeech fallback. For text-to-speech you can use ElevenLabs voices or the browser SpeechSynthesis engine (keyless), with per-message read-aloud buttons.
Keyless options like local Whisper and browser speech work without any API key, so voice is available out of the box.